Search papers, labs, and topics across Lattice.
This study investigates the necessity of high-degree-of-freedom adaptive attention mechanisms in traffic forecasting by comparing them to simpler global aggregation methods. Using a controlled ablation framework across six traffic benchmarks, the authors find that both uniform full-range mixing and standard spatial attention yield similar performance, with only a 0.14% difference in mean absolute error (MAE). The findings suggest that simpler methods can effectively capture global spatial information, challenging the prevailing reliance on complex attention mechanisms in this domain.
Global spatial information in traffic forecasting can be captured just as effectively with simpler methods, raising questions about the necessity of complex attention mechanisms.
Existing traffic forecasting models commonly focus on extracting spatial dependencies, particularly global spatial information, which characterizes the representations obtained through interactions between each individual node and all nodes across the traffic network. However, the underlying mechanism by which such global information is modeled and extracted remains insufficiently investigated. Whether global information must be extracted by high-degree-of-freedom adaptive attention or can be captured by a simple global aggregation operator remains unclear. For this purpose, we design a controlled ablation framework that replaces only the spatial mixing module to test attention-based global interaction. Across six traffic benchmarks, uniform full-range mixing and standard spatial attention each achieve lower MAE on three datasets, with only a 0.14% difference in mean MAE, while the former reduces node-scale spatial mixing complexity from O(N2) to O(N). Mechanism analysis further decomposes spatial attention into a row-uniform global background and a non-uniform residual. The residual shows dataset-dependent marginal value, suggesting that spatial attention should be justified by stable gains beyond a row-uniform global background. The corresponding source code is publicly available at: https://github.com/uuesti/U-Trans
